MASR: A Modular Accelerator for Sparse RNNs

TL;DR

MASR is a modular accelerator designed for sparse bidirectional RNNs, significantly improving energy efficiency and performance for speech recognition tasks by exploiting sparsity and dynamic activation optimizations.

Contribution

The paper introduces MASR, a novel modular architecture that accelerates sparse RNNs by exploiting sparsity in activations and weights, with dynamic optimizations for high efficiency.

Findings

MASR achieves 2x area reduction compared to state-of-the-art accelerators.

MASR provides 3x energy savings over existing solutions.

MASR delivers 1.6x performance improvements in RNN acceleration.

Abstract

Recurrent neural networks (RNNs) are becoming the de facto solution for speech recognition. RNNs exploit long-term temporal relationships in data by applying repeated, learned transformations. Unlike fully-connected (FC) layers with single vector matrix operations, RNN layers consist of hundreds of such operations chained over time. This poses challenges unique to RNNs that are not found in convolutional neural networks (CNNs) or FC models, namely large dynamic activation. In this paper we present MASR, a principled and modular architecture that accelerates bidirectional RNNs for on-chip ASR. MASR is designed to exploit sparsity in both dynamic activations and static weights. The architecture is enhanced by a series of dynamic activation optimizations that enable compact storage, ensure no energy is wasted computing null operations, and maintain high MAC utilization for highly parallel accelerator designs. In comparison to current state-of-the-art sparse neural network accelerators (e.g., EIE), MASR provides 2x area 3x energy, and 1.6x performance benefits. The modular nature of MASR enables designs that efficiently scale from resource-constrained low-power IoT applications to large-scale, highly parallel datacenter deployments.